Explore the brain
What the brain holds and the surfaces you can reach it through.
The Unison brain is a persistent memory layer that sits alongside your AI agents and tools. This page explains what it stores and how to reach it.
What the brain holds
Behind a single API live three interlocking stores:
- Documents — free-form text, conversations, code snippets, notes. Anything you ingest ends up here, chunked and vector-indexed for semantic recall.
- Entities — people, companies, projects, concepts. The brain extracts and links entities across documents so you can query by name or relationship, not just by keyword.
- Bitemporal facts — structured predicates attached to entities with both a valid-time and a record-time. Useful for things that change: job titles, pricing decisions, API versions.
Every recall call (GET /v1/brain/context) runs a planner that blends all three — hybrid search over documents, entity graph traversal, and fact lookup — and returns a single contextMd block ready to inject into a model prompt.
Surfaces
Reach the brain from wherever you already work:
| Surface | Package / link |
|---|---|
| TypeScript SDK | @unisonlabs/sdk |
| Python SDK | unisonlabs |
| CLI | @unisonlabs/cli |
| MCP server | @unisonlabs/mcp — plugs into Claude Code, Cursor, Codex |
| Filesystem mount | unison-fs — mount the brain as a local directory |
| Framework integrations | LangChain, LlamaIndex, Vercel AI SDK, Mastra, Pipecat — see Integrations |
| Raw HTTP | Brain API reference |
All surfaces share the same underlying brain: a memory written via the Python SDK is immediately recallable from the MCP server in Cursor.
Go deeper
- How recall and ingest work: Learn → Recall and Learn → Ingest
- Core concepts: How Unison works and What is a team brain?
- Get a key and make your first call: Create an API key → Add your first memory
Create an API key
Get a usk_live_ key for the Unison Brain API — via the dashboard, CLI, or the headless email-OTP API.
Build a coding agent with memory
A practical architecture for a coding agent that gets smarter every session: the recall → act → verify → write-back loop, what to store, and the code per surface — with what the research actually shows.